Directional Asymmetry of Long-Distance Dispersal And

Journal of Biogeography (J. Biogeogr.) (2005) 32, 741–754

SPECIAL Directional asymmetry of long-distance PAPER dispersal and colonization could mislead reconstructions of biogeography Lyn G. Cook* and Michael D. Crisp

School of Botany and Zoology, The ABSTRACT Australian National University, Canberra, Aim Phylogenies are increasingly being used to attempt to answer biogeographical Australia questions. However, a reliance on tree topology alone has emerged without consideration of earth processes or the biology of the organisms in question. Most ancestral-state optimization methods have inherent problems, including failure to take account of asymmetry, such as unequal probabilities of losses and gains, and the lack of use of independent cost estimates. Here we discuss what we perceive as shortcomings in most current tree-based biogeography interpretation methods and show that consideration of processes and their likelihoods can turn the conventional biogeographical interpretation on its head. Location Southern hemisphere focus but applicable world-wide. Methods The logic of existing methods is reviewed with respect to their adequacy in modelling processes such as geographical mode of speciation and likelihood of dispersal, including directional bias. Published reconstructions of dispersal of three plant taxa between Australia and New Zealand were re-analysed using standard parsimony and maximum likelihood (ML) methods with rate matrices to model expected asymmetry of dispersal. Results Few studies to date incorporate asymmetric dispersal rate matrices or question the simplistic assumption of equal costs. Even when they do, cost matrices typically are not derived independently of tree topology. Asymmetrical dispersal between Australia and New Zealand could be reconstructed using parsimony but not with ML. Main conclusions The inadequacy of current models has important consequences for our interpretation of southern hemisphere biogeography, particularly in relation to dispersal. For example, if repeated directional dispersals and colonization in the direction of prevailing winds have occurred, with intervening periods of speciation, then there is no need to infer dispersals against those winds. Failure to take account of directionality and other biases in reconstruction methods has implications beyond the simple misinterpretation of the biogeography of a taxonomic group, such as calibration of molecular clocks, the dating of vicariance events, and the prioritization of areas for conservation. Keywords *Correspondence: Lyn Cook, School of Botany Ancestral areas, Antarctic Circumpolar Current, Australia, directional dispersal, and Zoology, The Australian National University, Canberra, ACT 0200, Australia. maximum likelihood, maximum parsimony, New Zealand, southern hemi- E-mail: [email protected] sphere, trait optimization, West-wind Drift.

distributions of organisms (Raven, 1973; Wardle, 1978; INTRODUCTION McDowall, 2004; Thorne, 2004). Even taxa that appear to be It is commonly accepted that vicariance and long-distance poor dispersers have biogeographical patterns that require an dispersal (LDD) have both contributed to the current inference of dispersal (e.g. amphibians, Vences et al., 2004;

ª 2005 Blackwell Publishing Ltd www.blackwellpublishing.com/jbi doi:10.1111/j.1365-2699.2005.01261.x 741 L. G. Cook and M. D. Crisp ratites, Cooper et al., 2001; Haddrath & Baker, 2001; hebes, Australia and New Zealand (e.g. Nothofagus, Linder & Crisp, Wagstaff et al., 2002; Scaevola, Howarth et al., 2003). Examples 1995), in contrast to the conventional vicariance scenario where also include taxa, such as Nothofagus, where the recent New Zealand is sister to Australia + South America. Similarly, accepted paradigm had been one of Gondwanan vicariance ocean currents also appear to play a role in dispersal between (Craw, 1989; Linder & Crisp, 1995; Swenson et al., 2001). Australia and New Zealand. For example, sister relationships Repeated large-scale biogeographical patterns that are among some galaxiids probably reflect relatively recent dispersal ascribable to LDD indicate that LDD can be directional between the two regions, possibly on an older background of (Wardle, 1978; McDowall, 2002; Les et al., 2003; Munõz et al., Gondwanan vicariance (Waters et al., 2000a; McDowall, 2002). 2004; Sanmartıń & Ronquist, 2004), despite claims by some If directional LDDC can lead to congruent patterns across authors (e.g. Croizat et al., 1974; Nelson & Platnick, 1981; taxa (e.g. Munõz et al., 2004; Sanmartıń & Ronquist, 2004), Craw, 1982; Humphries, 2001; McCarthy, 2003) that LDD is then multiple occurrences of directional LDDC within a taxon inherently stochastic. Although successful LDD and establish- must also be a possibility. That is, if there are repeated patterns ment (hereafter referred to as long-distance dispersal and across distantly related taxa, why not among closely related colonization; LDDC) are conditional on variables such as the taxa? When dispersal is very common, the gene pools of the dispersability of the organism or its propagules, favourable populations either side of the semipermeable barrier remain in environmental conditions for dispersal and a suitable habitat contact and both populations will remain conspecific. How- for establishment, numerous processes involved in dispersal ever, if there is no further dispersal between the two have a strong directional component. populations after establishment then, over time, they will both Directional processes include dispersal from continental diverge and become separate species (allopatric speciation). landmasses onto newly created islands (Bush & Whittaker, The possible role of repeated directional dispersal and its 1991; Wagner & Funk, 1995; Nepokroeff et al., 2003), down effect on phylogeny-based reconstructions has been largely river systems (e.g. Campbell et al., 2002; Levine, 2003) and, neglected. Directionality has significant implications for bio- importantly for global-scale patterns, with prevailing winds geographical reconstructions derived from phylogenies of taxa (e.g. Pedgley, 1982; Greathead, 1990; Levin et al., 2003) or that span areas where LDDC is likely to have been directional. ocean currents (e.g. Wares et al., 2001; Calsbeek & Smith, By not considering all alternatives, reconstructions may appear 2003; Gaines et al., 2003). Additionally, animal-assisted dis- more decisive than is perhaps warranted. persal, such as by birds, might also be directional (e.g. Wardle, 1978; Wenny & Levey, 1998). INTERPRETING TREES FOR BIOGEOGRAPHY Wind and ocean currents are probably the two major abiotic processes that might lead to congruent LDDC patterns across Most biogeographical studies typically substitute the terminal multiple taxa (e.g. Raven, 1973; Wardle, 1978; Gaines et al., names of a species or gene tree with the geographical areas of 2003; McDowall, 2004; Munõz et al., 2004). In the Southern each taxon to produce a taxon–area phylogeny. If a species tree is Hemisphere, for example, currents are dominated by the used, each node of the phylogeny represents a speciation event. Antarctic Circumpolar Current (ACC) and West-wind Drift Speciation falls into two main categories – allopatric (speciation (WWD) (Bowler, 1982; Colls & Whitaker, 2001) which were in geographical isolation) and sympatric (reproductive isolation initiated after the rifting of South America and Australia from arising despite individuals’ continuously having an opportunity Antarctica c. 38 Ma (Veevers et al., 1991). These currents may be to interbreed). Allopatric speciation is probably more common responsible for the close taxonomic affinities between southern than sympatric speciation (e.g. Barraclough & Nee, 2001; Turelli hemisphere localities that are in conflict with conventional et al., 2001) and, to date, there are few examples where sympatric continental drift patterns (e.g. Raven, 1973; Waters et al., 2000a; speciation is supported empirically. Interpretation of causes of McDowall, 2002; Munõz et al., 2004; Sanmartıń & Ronquist, the speciation events, in the context of tree topology, is used to 2004). generate or discriminate among alternative biogeographical The westerly wind flow is thought to be responsible for strong hypotheses. In a phylogeny of organisms, taxa diverging at a affinities between the New Zealand and Australian flora, with terminal node (i.e. ‘twigs’) are typically interpreted as ‘the most some of the New Zealand flora being derived from propagules derived’ or ‘most recently diverged’ taxa (e.g. taxa A and D in blown east across the Tasman Sea from Australia, currently a Fig. 1). In terms of biogeography, the equivalent would be ‘most distance of c. 2000 km (Raven, 1973; Wardle, 1978; Macphail, recent dispersal to’ or ‘most recent vicariance event’ (Mast & 1997; Jordan, 2001; Pole, 2001). Shared extant species across this Givnish, 2002; Winkworth et al., 2002a). gap suggest that the process is ongoing, although not universal, Inference of ancestral areas using a taxon–area phylogeny and a number of traits apparently influence the likelihood of typically depends on maximum parsimony (Bremer, 1992; success (Wardle, 1978; Jordan, 2001; McGlone et al., 2001). For Maddison & Maddison, 1992; Ronquist, 2003), maximum example, populations with very light, wind-dispersed spores, likelihood (ML) (Cunningham et al., 1998; Cunningham, such as ferns, appear more likely to maintain genetic contact 1999; Mooers & Schluter, 1999; Pagel, 1999a; Lutzoni et al., over large oceanic gaps and remain conspecific over long 2001) or, more recently, Bayesian (Huelsenbeck & Bollback, periods (Ranker et al., 1994; Wolf et al., 2001). In a phylo- 2001; Pagel et al., 2004; Ronquist, 2004) approaches to genetic context, this results in a sister relationship between reconstruction of ancestral states at internal nodes of a tree.

742 Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd Asymmetrical dispersal and biogeography

(a) (b) events, such as vicariance, dispersal, sympatric speciation (duplication) and extinction, having varying probabilities Area Y Y Y Z Y Y Y Z (Ronquist, 1998; Sanmartı´n & Ronquist, 2004). DIVA uses a Taxon BC D A BC D A cost matrix to assign a cost to each event type and minimizes the total cost of optimizing areas over the tree (Ronquist, 1997). Speciation by vicariance is considered the null expectation and is given zero cost. Sympatric speciation is also given zero cost. Extinction and dispersal each have unit cost. Therefore, DIVA tends to favour vicariance reconstructions over dispersal. In Fig. 1, DIVA infers a dispersal of the ancestor of D + A into area Z (cost ¼ 1) followed by vicariance (cost ¼ 0) (cf. Ronquist, Figure 1 Contrasting interpretations of an area. Grey indicates the 1997, Fig. 3a). The user of DIVA may choose to restrict the source population through time and black indicates dispersed breadth of possible ancestral distributions to fewer than the sum populations. (a) Fitch parsimony explains only the terminal node, by a jump dispersal from area Y resulting in a new species (A) in area of descendent areas, and that may be a single area. A broad Z. Vicariance explains the same node as a split between areas Y and Z, ancestral distribution would be more consistent with a history of resulting in two new species, respectively D and A. Neither inter- vicariance, and a narrow one with a history of dispersal. pretation explains the earlier, apparently sympatric speciation events leading to species B and C in area Y. Therefore, these events Maximum likelihood are not costed by vicariance or Fitch models. (b) Repeated jump dispersals into area Y from a source population (A)inZ give rise to Maximum likelihood modelling of the evolution of traits on species B-D, so that species A (and its area Z) is placed at a terminal trees has several advantages over parsimony (Pagel, 1994, node by cladistic analysis. This interpretation explains all nodes and 1999b; Cunningham, 1999; Mooers & Schluter, 1999; Nielsen, events and, although it may seem unparsimonious compared with 2002) and has been used to estimate ancestral areas (e.g. Fitch or vicariance models, its overall cost is equivalent. Nepokroeff et al., 2003). Whereas parsimony reconstructs a single state at each node, ML can indicate the probabilities of Parsimony alternative states. If change is highly probable relative to Fitch optimization (equally weighted parsimony) is a com- branch length, then a basic assumption of parsimony is monly used method (e.g. Enghoff, 1995; Andersson & Chase, violated and the rate of evolutionary change is likely to be 2001; Manos & Stanford, 2001; Lieberman, 2002) that, in underestimated. In contrast, ML uses branch lengths to model effect, models dispersal and allopatric speciation (Ronquist, the likelihood of change along each branch. An ML analysis of 2003). Only dispersal events are counted and, whilst sympatric the tree in Figs 1 and 2 (pectinate ultrametric tree with internal speciation is allowed, it is not counted in the parsimony sum nodes of equal length), using a continuous time Markov model (Ronquist, 2003). This involves an a priori assumption that is of trait evolution implemented by the program discrete applied across the tree and takes no account of differences (Pagel, 1994, 1999a, http://www.ams.reading.ac.uk/zoology/ among taxa, such as mode of speciation. In Fig. 1, Fitch pagel/), also favours area Y as the more likely ancestral area at parsimony explains only the terminal node, by long-distance internal nodes of the tree (likelihood at t3 ¼ 0.70, t5 ¼ 0.69). dispersal from area Y resulting in a new species (A) in area Z. This leads to the conclusion that a population in area Y was the Bayesian inference source and that dispersal occurred only once – at the terminal grey node (Fig. 1a). It is assumed that dispersal occurred at the Bayesian inference has the advantage over both parsimony and time of this speciation event, and in the direction of the single ML methods of trait reconstruction of taking phylogenetic nested area (Z). uncertainty, such as topology, parameter values, and branch Subtree analysis (Nelson & Ladiges, 1996; Ebach & length estimations, into account (Huelsenbeck & Bollback, Humphries, 2002) uses an event-based approach that explicitly 2001; Huelsenbeck et al., 2003; Pagel et al., 2004; Ronquist, favours vicariance over dispersal (Page, 1994; Nelson & 2004). However, it has the same problem as other methods of Ladiges, 2001; Ronquist, 2003). Using this method, the ﬁnal not handling asymmetry. speciation event leading to A and D (Fig. 1) would be treated as a vicariance event between areas Z and Y, whereas taxa B ALTERNATIVE EXPLANATIONS and C, also occurring in Y, would be considered paralogous (relative to D) and discarded from further consideration. We argue that methods currently used for biogeographical Both Fitch and Subtree Analysis are intuitively unsatisfac- reconstructions are unsatisfactory because they do not consi- tory because they fail to explain the number of species in the der all nodes, typically treat process explanations as independ- paraphyletic residual in area Y (B ) D, Fig. 1). ent of each other, and/or fail to take account of likely repeated All nodes are taken into account in a process-based approach patterns such as those induced by directionality. Directional to biogeography (Ronquist, 1994, 1998, 2003), such as DIVA, bias can mislead all tree-based reconstructions of trait which uses parsimony to choose between different kinds of evolution (Mooers & Schluter, 1999; Omland, 1999; Pagel,

Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd 743 L. G. Cook and M. D. Crisp

Area through sympatric speciation or arisen in allopatry and then Time Z Y Area Z dispersed into the common area. An interpretation of symp- Taxon Time A0 atric speciation from the literature requires caution, however,

t 0 A0 because taxa grouped under one area by some authors may occupy discrete non-overlapping areas within that larger area and hence not be sympatric in the strict sense. Because biogeography is process-based, all nodes in a Area Y Z topology need an explanation. In parsimony analysis, the total

t 1 A1 Taxon B 2 A2 is minimized over all variables and may necessitate unparsimonious scores for some individual components (Farris, t 1 1983). Therefore all events, including both sympatric and allopatric speciation, should be factored into the parsimony

t 2 A2 B equation. Therefore, simply (equation 1):

total nodes ¼ sympatric speciationþ B3 Area Y Y Z allopatric speciation (via vicariance or dispersal) extinction t 3 A3

Taxon B 4 C 4 A4 i.e. NT ¼ S þ V þ D E

t 3 There are three nodes in Fig. 1, requiring a minimum of three B4 t 1 explanations (events) for the apparent divergence and distri- t 4 A4 C4 bution of taxa.

Non-independence of variables

B5 Analyses of taxon–area phylogenies are often performed Area Y Y Y Z C5 t 5 A5 to optimize one process, in isolation from others, in order to Taxon B 6 C 6 D 6 A6 t 6 answer questions such as ‘how many dispersals are required to explain the current distribution of this lineage?’ However, t 5 dispersals are not independent of the other explanations for B6 t 3

t 6 6 divergence events (nodes). Allopatric speciation (dispersal or A C6 t 1 D6 vicariance) and sympatric speciation cannot occur for the same taxon simultaneously and are thus not independent of one another for explaining a node. That is, if dispersal is inferred to Figure 2 Stepwise sequence of dispersal events leading to the phylogeny in Fig. 1b. Boxes indicate areas and species occurring explain a node, then vicariance cannot simultaneously explain within them. Arrows show dispersal events. In the right-hand the same node. Similarly, if dispersal is not inferred to explain column, the area cladogram is shown as it develops through time. a node, another explanation (such as sympatric speciation or Shading and symbols for areas and species are as in Fig. 1. Changes vicariance) is required. As pointed out above, choosing an in subscripts of species indicate anagenetic change. Timing of option that minimizes the number of inferred dispersals events is indicated by t0–t6. minimizes one of the possible process explanations for nodes but it may not minimize overall parsimony for a topology 1999b; Ree & Donoghue, 1999; Oakley & Cunningham, 2000), because dispersal, in some cases, will explain only some nodes including ancestral areas. and other processes will need to be invoked to explain the others. Counting all nodes Repeated directional patterns and processes Biogeographical problems are concerned with process, typically involving discrimination between vicariance and dis- The simplistic assumption of assigning equal probability to persal, and are hence event-based (Ronquist & Nylin, 1990). different kinds of evolutionary events (e.g. gains and losses) Optimizing dispersal or vicariance requires consideration of may often be unfounded (Omland, 1997; Schluter et al., 1997; sympatric speciation (duplication), because co-occurrence of Cunningham et al., 1998; Pagel, 1999b; Nielsen, 2002; related taxa requires either speciation in situ or speciation in Ronquist, 2003; Felsenstein, 2004). For example, multiple allopatry with subsequent range expansion (dispersal) to losses of features under parallel selection appears to be a reconnect the two taxa (Ronquist, 1998). If speciation occurred common phenomenon in evolution, with some well-docu- in allopatry, vicariance or dispersal explanations are required, mented examples. Repeated loss or reduction of legs and rather than implying sympatric speciation. For example, the antennae is associated with a galling habit in scale insects paraphyletic residuals in area Y (Fig. 1) could each have arisen (Cook & Gullan, 2004). Flight appears to have been lost many

744 Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd Asymmetrical dispersal and biogeography times in different lineages of pterygote insects (Whiting et al., frequent, the disjunct populations will function as a single gene 2003), especially in historically stable, isolated (island-like) pool. However, if the period of time between dispersal/ habitats (Wagner & Liebherr, 1992). Flightless species of birds establishment events (t2 ) t1) is sufficient, a subpopulation also appear to have evolved repeatedly and independently established in area Y may diverge from sp. A to become a following dispersal of their volant ancestors to oceanic islands different species (sp. B). We now have a sister relationship of sp. (Trewick, 1997; Slikas et al., 2002). Intuitively and empirically, A in area Z and sp. B in area Y. Both populations will undergo loss of flight appears to be much more common, and easier in anagenetic change, which is shown by their changing subscripts an evolutionary sense, than evolution of flight. through time in Fig. 2 (e.g. from A1 to A2 over this period).

Treating each process (or trait change, or losses and gains) A later dispersal event (t3) in the same direction as the ﬁrst as being equally likely, when they are not, can have a signiﬁcant may lead to a colonization of area Y by sp. A3. Because sp. A3 is effect on trait reconstruction using current methods, such as distinct from its sister sp. B3, two species now occur in Y (A3 parsimony and ML (e.g. Omland, 1997, 1999; Pagel, 1999b; and B3). Again, there may be differentiation between the

Schultz & Churchill, 1999; Oakley & Cunningham, 2000; populations of sp. A3 in areas Z and Y leading to isolation of

Huelsenbeck et al., 2002; Nielsen, 2002; Ronquist, 2003). In the colonizing population in area Y, which becomes species C4 biogeography, the equivalent to equal rates of losses and gains (t4). Periodic episodes of colonization of area Y from area Z is the treatment of a dispersal from Y to Z as being equally (with differentiation of populations between events) could lead likely as a dispersal from Z to Y (Fig. 1a). The assumption does to multiple species in area Y, whilst leaving a single species in not consider the inferred process of dispersal. Whilst dispersal area Z (A6 at t6). Thus, on a species phylogeny, the area may be rare over time, it may be non-random in space – as occupied by the original species (Z) would be nested within a shown by the examples cited above. If there is directionality to paraphyletic area Y. Species A in Z will be attached to a dispersal, it would be expected that there would be a high terminal node if it undergoes anagenetic change, as expected, frequency for a single dispersal per taxon in that direction, a with character ﬁxation between episodes of dispersal to and lower frequency of two dispersals per taxon in that direction, speciation in area Y. Hence it will share apomorphies with the and lower still for three. However, depending on the strength successively dispersed new species and will appear terminal in a of the directionality, the frequency of multiple dispersals may phylogeny. Change in a population is inevitable over time, but be higher than for a single dispersal in the opposite direction: if dispersal events are frequent relative to allopatric speciation the greater the strength of directionality, the higher the events, then the organismal phylogeny will be a polytomy, frequency of multiple dispersals in one direction relative to rather than resolved with a ‘nested ancestral area’. dispersal in the opposite direction (Fig. 5). The NAA alternative explanation is as parsimonious as the The source taxon of dispersing organisms may remain a usual interpretation if the parsimony sum is considered single interbreeding population persisting over a long period, (equation 1 above). That is (S + V + D ) E ¼ NT): i.e. a cohesive gene pool changing through time, while Usual interpretation: 2 + 0 + 1 ) 0 ¼ 3 (dispersal inter- spawning successive new populations into another area pretation), or 2 + 1 + 0 ) 0 ¼ 3 (vicariance interpretation). (a sink). If dispersal is non-random, this may result in a NAA interpretation: 0 + 0 + 3 ) 0 ¼ 3 (all dispersal), or ‘nested ancestral area’ (NAA) – a taxon from the ancestral area 0+1+2) 0 ¼ 3 (if one vicariance event e.g. Fig. 3). appears nested within a paraphyletic group of taxa occurring in If dispersal from Z to Y is much more likely than dispersal the sink area. That is, the ancestral area will appear at a from Y to Z, then the NAA explanation will also be the most terminal node in a taxon–area cladogram and is typically likely explanation. interpreted as ‘recently colonized’. IMPLICATIONS OF A NESTED ANCESTRAL AREA HOW TO BECOME A NESTED ANCESTRAL Failure to take account of directionality and other biases in AREA: AN ALTERNATIVE EXPLANATION FOR reconstruction methods has implications beyond the simple FIG. 1 misinterpretation of the biogeography of a taxonomic group. As explained above, the commonly used Fitch optimization Several are outlined below. interprets the position of species A in Fig. 1 as a recent dispersal from area Y to Z. A converse interpretation is possible if Calibration of molecular dates, and their repetitive events occur. For simplicity, extinction is assumed to interpretation be absent in the following reconstruction of such events (Fig. 2).

At time 0 (t0), species A occurs in area Z. Species A usually There is no known fossil record for most taxa and it is very disperses locally but there may be occasional LDDC to area Y incomplete for others. In addition, much of the plant fossil

(t1). Establishment of sp. A in Y depends on the process of getting record consists of pollen and spores that are seldom identi- to Y (e.g. wind, storms), and population establishment once in Y ﬁable to species level and often provide little resolution within (e.g. habitat it arrives in, numbers and sex of individuals). Either species-diverse groups, e.g. eucalypts (Hill, 1994). The rela- dispersal from area Z to area Y may be unusual (e.g. exceptional tionship between pollen taxa and species-level diversity is storm events) or establishment events may be rare. If events are unclear, even in well-studied taxa that have a good record of

Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd 745 L. G. Cook and M. D. Crisp

Area Area X (a) (b) X Area Y Y Y Z Y Y Y Z Taxon M Taxon D BC D A 6 BC A Time t t 6 t 0 M0 t 5 40 Ma t 5

t 3 t 3 Z Y t 1 t 1 40 Ma Area Y Z

t 1 1 1 M M Taxon B 2 A 2 (c) (d) Vicariance t 1 Area Y Y Y Z Y Y Y Z M Taxon D BC D A 6 BC A t t 6

t 2 A2 B 2 t 5 10 Ma t 5

t 3 t 3

t 1 t 1 40 Ma

B 3 Area YY Z t 3 A3

Taxon B 4 C 4 A4 Figure 4 Implications of molecular dating of nodes under different interpretations of the taxon–area phylogeny. (a)–(b) assu- t 3 ming that the Z-Y split is used to calibrate the clock, e.g. a

B 4 Vicariance t 1 vicariance event at 40 Ma. (a) In the standard interpretation, this t 4 A4 C 4 is also the date of the split between species A and D hence node t5 dates at 40 Ma and node t1 much earlier. (b) In the nested ancestor interpretation, the vicariance event may have occurred much earlier, implying a younger age for all the speciation events, e.g. B 5 Area YY Y Z t 5 C 5 t1 ¼ 40 Ma. (c)–(d) assuming independent calibration of the

A5 Taxon B 6 C 6 D 6 A6 t 6 molecular clock, i.e. a node outside the ingroup. (c) In the standard interpretation, the split between areas Y and Z is assumed t 5 to coincide with the split between species A and D, i.e. 10 Ma. No

t 3 B 6 fossils of the entire lineage are expected in Z before this date. t 6 A6 Vicariance C 6 t 1 (d) In the nested ancestor interpretation, speciation event A-D D 6 represents only the latest such split between Y and Z, with the earliest having occurred much earlier (e.g. at 40 Ma). Fossils of the Figure 3 Stepwise sequence of vicariance followed by dispersal lineage are predicted in Z from 40 Ma. events leading to the phylogeny in Fig. 1b. Boxes indicate areas and species occurring within them. Arrows show dispersal events. In the right-hand column, the area cladogram is shown as it would typically be assigned to the terminal node. This results develops through time. Shading and symbols for areas and species from a belief that the fossil could not be older because the are as in Fig. 1. Changes in subscripts of species indicate ana- group was not in Z until the most recent divergence event. If genetic change. At time t0, species M occurs in area X which the fossil is placed at the most recent node (t5) (e.g. Fig. 4c), vicariates at t1 to form areas Z and Y. Taxon M speciates in allopatry to form two daughter species: A and B. Two dispersal then deeper nodes in the tree will be estimated to have older events from area Z to area Y follow, with sufﬁcient time between divergence times than if it is placed in the position inferred each for the populations to have speciated. The single taxon in area from a NAA interpretation at t1 (Fig. 4d). Z undergoes anagenetic change through time. Although a popu- A NAA situation also has implications for discriminating lation has been present in area Z for the same length of time as it between a vicariance and all-dispersal interpretation. For has been in area Y, conventional reading of the tree is such that example, a common question in biogeography is whether a area Z appears only recently colonized. vicariance event between Y and Z may explain part of the distribution of a lineage. Hence, the timing of nodes is increasingly being used to discriminate between vicariance and both pollen and macrofossils, e.g. Nothofagus (Hill, 2001). This dispersal (e.g. Swenson & Bremer, 1997; Renner et al., 2000, uncertainty in the placement of fossils on a phylogeny means 2001; Winkworth et al., 2002a; Givnish & Renner, 2004). If, that a failure to recognize a NAA situation has severe under a conventional reading of Fig. 1, a molecular date suggests consequences for the calibration of molecular clocks. For that the most terminal node is too recent to be explained by example, if a dated fossil of the lineage is known from area Z vicariance then a dispersal explanation is typically inferred (e.g. and is used as a calibration point for molecular dating, Leys et al., 2000; Trewick, 2000). However, under alternative uncertainty in its placement could cause large differences in biogeographical reconstructions for Fig. 1, another node may be estimates for other nodes. A fossil found in Z that is assignable the result of a vicariance event (e.g. Fig. 3). Thus, if the basal to the group, but not to an individual species within the group, node in Fig. 1 had an estimated timing consistent with a

746 Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd Asymmetrical dispersal and biogeography

geological record may indicate that one of two terminal sister areas is continental land and much older than its sister area, a recent volcanic island. Dating of nodes, in this case, may rule High A to B out possibilities (Hunn & Upchurch, 2001; Donoghue & Moore, 2003; Page, 2003). If the common node of sister taxa, B to A one on the island and one on the mainland, is about the same age as the island, then it seems most likely that dispersal occurred from the continent to the island, and not vice versa. Under the conventional late dispersal (Fitch, ML) interpret- Frequency ation of Fig. 1, it is assumed that the lineage including species A-D was absent in area Z until the ﬁnal speciation and Low dispersal event. Presence of fossils of that lineage in area Z older than the date of the most recent node (grey node, Fig. 1) would falsify the late dispersal hypothesis. By contrast, the 1234 NAA hypothesis assumes a continuous presence of the lineage

Number of dispersals in area Z from t0 to t6. Fossils in Z from any time before t5 would support this hypothesis, and a continuous record from Figure 5 Directional dispersal. The frequency of multiple dis- t to t would strongly support it. Therefore, a fossil record that persals is likely to be lower than that for a single dispersal. 0 6 Asymmetry of dispersal arises when dispersal in one direction is older than predicted, and in conﬂict with conventionally (e.g. from A to B) is more frequent than dispersal in the other read molecule-based trees, should lead to a reassessment of tree direction (B to A). In this example, it would be more likely to interpretation. The NAA model could explain the discrepancy observe three dispersals from A to B than to see one dispersal from between molecular and fossil estimates of the time of B to A (dotted line). divergence because it predicts a longer history of the lineage in the NAA area than that determined from the most recent node on the tree. In the case of such a re-interpretation, the vicariance event, then vicariance should still be considered molecular clock estimates would need a calibration point possible for explaining that part of the biogeography of the outside the lineage in question. lineage. The parsimony cost for Fig. 1 would still be three if there was an early vicariance event between areas Y and Z, followed by Independent estimates of asymmetry two dispersals from Z to Y. Asymmetry (directionality) can be built into character-state reconstruction methods to try to provide more realistic Conservation priorities reconstructions of ancestral states. Approaches for doing this It has been argued that centres of origin, such as source include step matrices (parsimony) (e.g. Ree & Donoghue, populations, should be given higher conservation priority 1998; Belshaw & Quicke, 2002), rate matrices (ML) (e.g. because such areas have spawned species in the past and Mooers & Schluter, 1999; Pagel, 1999a; Oakley & Cunning- may therefore be likely to do so in the future (e.g. Fjeldsa, ham, 2002; Nepokroeff et al., 2003), or Bayesian prior 1994; Linder, 1995; Soltis & Gitzendanner, 1999). Under probabilities (e.g. Schultz & Churchill, 1999). A way of using conventional ancestral area reconstruction, area Y in Fig. 1 is these methods is to ask the question ‘How biased does reconstructed as the ancestral area, whereas under a NAA asymmetry need to be to give an alternative reconstruction of scenario area Z is the ancestral area. Clearly this has ancestral states?’ (e.g. Ree & Donoghue, 1998) and then to implications for conservation management. The two opposing make a decision about whether the bias required is extreme hypotheses suggest different assumptions about the evolutionarily realistic. This approach has been used for population histories of the taxa in those areas. Under a morphology (e.g. loss or gain of wings in stick insects, Whiting conventional reading, area Y might have had an unstable et al., 2003) and behaviour (e.g. life history strategy in history in leading to speciation among local populations. In ichneumonoids, Belshaw & Quicke, 2002). contrast, a NAA hypothesis suggests that area Z has had a long For Fig. 1, the reversal in ancestral area reconstruction, from period of stability, with taxon A persisting through to the one dispersal from area Y to area Z to three dispersals from current time. area Z to area Y, occurs when the asymmetry is coded to be > 3 : 1 (Fitch reconstruction, deltran and acctran,as implemented in paup* ver. 4.0b10, Swofford, 2002). If all TESTING HYPOTHESES – POSSIBLE SOLUTIONS processes are considered, and all nodes counted (equation 1), the reversal in reconstruction can occur whenever dispersal in Geological and fossil record one direction is down-weighted relative to dispersal in other A good geological record can enable choice between compet- directions and sympatric speciation (i.e. parsimony sum is ing dispersal hypotheses (e.g. Fig. 4a,b). For example, the least for three dispersals).

Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd 747 L. G. Cook and M. D. Crisp

The problem is that in most circumstances there is no exhibit only a proportion of the genetic diversity of the objective way to decide whether the required asymmetry is parental population (Templeton, 1998). This is because they realistic. It has been suggested that asymmetry could be have been sampled from the parental population (via estimated using comparisons across other taxa (Omland, 1999 dispersal) – only some individuals from the parental popu- and citations therein). However, as acknowledged by Omland lation contribute to the daughter population. Thus, the (1999), estimates derived from phylogenies have the problem relationship between two populations in different areas could of circularity in that they use tree-reading to determine be considered when deciding which represented the putatively character transitions and therefore can be biased as a result of dispersed population. For example, consider terminals A and the same asymmetry that is being estimated (Cunningham, D in Fig. 1 to represent two populations of the one species. If 1999; Mooers & Schluter, 1999; Omland, 1999). That is, the population A comprised genetic diversity greater than, and estimates are not independent of tree-reading problems and encompassing, that of population D, we could be more are therefore subject to the same problem of underestimation confident that A represented the ancestral population, at least of asymmetry. To avoid the circularity of phylogeny-based for population D. asymmetry estimates, parameter values need to be derived Dean & Ballard (2004) provide an empirical example of independently of tree-interpretation, such as from biology or reconstruction of ancestral area using population genetics and other known processes that can be estimated. This has not yet phylogenetics for Drosophila simulans. Similar studies provide been attempted in biogeography. evidence for directional migration among populations, particularly for marine invertebrates, including sea-stars (Waters & Roy, 2004) and barnacles (Wares et al., 2001). Multiple Possible sources of independent estimates of asymmetry invasions from stable source populations into new or unstable Sanmartıń & Ronquist (2004) estimate that 6.59% of dispersals habitats have also been inferred from population genetic among Southern Hemisphere taxa were from Australia to New studies. Three-spined sticklebacks are thought to have repeat- Zealand, whereas only 3.82% were from New Zealand to edly colonized and speciated into fresh-water lakes from a Australia. That is, dispersal with the WWD was almost twice as marine source (Gasterosteus aculeatus) in the Strait of Georgia, common as dispersals against it. These values are likely to be British Columbia, when sea levels rose twice during the underestimates of the real asymmetry because Sanmartıń& Pleistocene (Taylor et al., 1997; Taylor & McPhail, 2000; Ronquist’s estimates are based on reconstruction of character Schluter et al., 2001). state changes (direction of dispersal) on phylogeny alone and The phylogeographical pattern that would be expected are probably biased as a result of the asymmetry itself. during the early stages of directional dispersal leading to a Interestingly, Sanmartıń & Ronquist (2004) determined that NAA situation has been recorded for Galaxias maculatus the observed number of dispersals was significantly less (Waters et al., 2000b). This fish occurs in coastal waters of frequent than that expected, whereas inferred duplications Australia (Tasmania), New Zealand and Chile. The three (sympatric speciation) were more frequent than expected. This regions form discrete haplotype clusters, with the exception of is the outcome that would be expected if there were multiple a haplotype from one individual from New Zealand that clearly directional LDDCs but with trees read in the conventional clusters with the Tasmanian group. The divergence between manner, inferring single dispersals (forcing dispersals towards the Tasmanian and New Zealand clusters suggest that gene the terminal nodes of trees) and thereby requiring greater flow is not common, but that a rare dispersal event (only one numbers of duplications to explain internal nodes. We suggest individual sampled) may have occurred from Tasmania to that the significant bias found by Sanmartıń & Ronquist (2004) New Zealand in the direction of the west to east ocean currents is consistent with the failure to recognize NAA situations. (Waters et al., 2000b). However, because Sanmartıń & Ronquist’s (2004) estimates Unfortunately, a genetic diversity argument cannot be are likely to be underestimates, a ratio of c. 2 : 1 could be used applied unambiguously to taxa that are no longer part of the as a conservative starting point to assess the possibility of one gene pool (i.e. different species). Gene pools diverge multiple dispersals. That is, reconstruction of dispersal through time and different population sizes, gene coalescence between Australia and New Zealand that can be reversed with times and selection pressures will affect genetic diversity. For a weighting scheme of 2 : 1 should favour a ‘multiple dispersal example, the original source population could undergo a out of Australia’ hypothesis rather than an a ‘single dispersal severe bottleneck and lose diversity such that it now appears out of New Zealand’ hypothesis. This minimum ratio method less diverse than its daughter. That is, over time, the differing could be extended to other areas where asymmetry is expected. evolutionary pathways of separate species will obfuscate population histories. To avoid these problems, an alternative use of population Genetic diversity genetics and phylogeography could be to examine extant Population genetics can give an indication of direction of species that are shared across areas of interest. An estimate of dispersal and could be used to derive an independent estimate direction of dispersal could be determined across populations of asymmetry. Populations that have been established follow- of multiple taxa and used to derive an estimate of asymmetry ing dispersal from a parental population are expected to unbiased by tree-reading. For example, if 90 of 100 species

748 Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd Asymmetrical dispersal and biogeography shared between two areas, such as Y and Z, were determined to Myosotis), none are inferred to have gone in the expected have originated in Z, then an asymmetric weighting scheme of eastwardly direction. We do not suggest here that dispersal 9 : 1 could be built into ancestral area reconstruction for a could not have been against prevailing winds, and the authors higher level group occurring in these two areas. This method give plausible scenarios for such dispersal. Rather, we use the assumes that conditions leading to the observed asymmetry phylogenies presented by Winkworth et al. (2002a) because were also present at the time of divergence of taxa in the focal they represent a recent study and provide a framework to group. Testing this assumption requires additional informa- demonstrate that alternative explanations could be considered. tion, for example, from climatic reconstructions for the period Phylograms were captured from original publications in question. Additionally, the biology of different organisms (Lockhart et al., 2001; Wagstaff et al., 2002; Winkworth et al., varies and therefore the causes and magnitude of asymmetry 2002b) using TreeThief v1.0 (Rambaut, 1999, http://evolve. will vary among taxa. Account could be taken of this in zoo.ox.ac.uk/software.html?id¼treethief). To determine the estimating asymmetry between regions and using species with level of asymmetry required to reverse the ‘out of New similar life-history traits to the taxon of interest. Zealand’ ﬁndings, we used Fitch parsimony (acctran and Australia and New Zealand currently share many species deltran), as implemented in paup* ver. 4.0b10 (Swofford, (e.g. Jordan, 2001; McGlone et al., 2001; Mummenhoff et al., 2002), and ML using a continuous time Markovian model 2004), providing a rich resource for estimating asymmetry in (Pagel, 1999a), as implemented in discrete (Pagel, 2000) and dispersal between the two regions if population level studies multistate (Pagel, 2003, http://www.ams.reading.ac.uk/zool- were to be conducted. ogy/pagel/).

AN EXAMPLE OF A POTENTIAL NESTED Parsimony ANCESTRAL AREA SCENARIO In all three data sets, unweighted Fitch parsimony (acctran and Australia and New Zealand provide a situation conducive to deltran) reconstructed New Zealand as the ancestral area at the development of NAA patterns. Although the two land all nodes containing Australian taxa, thereby inferring dispersal masses first separated 85 Ma (Kroenke, 1996; Lee et al., 2001), from NZ to Australia. We applied asymmetry, in favour of when New Zealand rifted from Antarctica, they are currently west to east (i.e. Australia to NZ), to the direction of inferred separated by only 2000 km of ocean – a distance that has been dispersal by using asymmetrical step-matrices with weight relatively stable for 55–60 Ma (Pole, 1994; Kroenke, 1996). ratios ranging from 2 : 1 to 7 : 1. A change in inferred Additionally, the two regions extend into the southern direction of dispersal, from a New Zealand ancestor to a latitudes of the Southern Hemisphere dominated by the ACC reconstruction of an equivocal or Australian ancestor, and WWD. It would therefore be predicted that dispersal occurred with a ratio of 2 : 1 for hebes, 4 : 1 for Ranunculus, should most often occur in the direction of the WWD from and 6 : 1 for Myosotis (Table 1). In other words, if dispersal Australia to New Zealand, and this has been postulated as the from Australia to New Zealand is more than six times more cause of the high number of shared species between the two countries (e.g. Raven, 1973; Jordan, 2001; McGlone et al., Table 1 Fitch-parsimony estimates of ancestral areas in examples used by Winkworth et al. (2002a). Asymmetrical rate matrices 2001). (indicated) were applied to dispersal direction, favouring Australia However, it has also been suggested that dispersal has to New Zealand occurred against the prevailing winds and ocean currents – from New Zealand to Australia (e.g. Wardle, 1978; Swenson & Hebes Ranunculus Myosotis Bremer, 1997; Wagstaff et al., 2002; Winkworth et al., 2002a; Sanmartıń & Ronquist, 2004). The recent reconstructions have Basal node; 1 : 1 NZ NZ NZ Basal node; 2 : 1 Aus or NH NZ NZ been based on the finding of Australian taxa nested within New Basal node; 3 : 1 Aus NZ NZ Zealand groups and, implicitly, an assumption of equal Basal node; 4 : 1 Aus Aus or NZ NZ probability of dispersal in all directions, i.e. random dispersal. Basal node; 5 : 1 Aus Aus NZ In the above cases, it is acknowledged that this is against the Basal node; 6 : 1 Aus Aus Aus (a), direction expected and authors have invoked unknown NH (d) alternative processes to explain the apparent anomaly. Basal node; 7 : 1 Aus Aus Aus We reanalysed the three data sets that contain taxa with Other Aus-NZ nodes; 1 : 1 NZ NZ – distributions including both Australia and New Zealand that Other Aus-NZ nodes; 2 : 1 NZ NZ – were used by Winkworth et al. (2002a) to support their ‘out of Other Aus-NZ nodes; 3 : 1 Aus NZ – New Zealand’ hypothesis. Interestingly, although the expected Other Aus-NZ nodes; 4 : 1 Aus Aus or NZ – direction of dispersal is, as acknowledged by Winkworth et al., Other Aus-NZ nodes; 5 : 1 Aus Aus – Other Aus-NZ nodes; 6 : 1 Aus Aus – with the westerly winds from Australia to New Zealand, no dispersal in this direction is inferred in any of the three data Results were the same for acctran (a) and deltran (d) reconstruc- sets. That is, despite four or five inferred dispersals from NZ to tions, except where indicated. Australia (two in Ranunculus; one or two in hebes; one in Aus, Australia; NZ, New Zealand; NH, Northern Hemisphere.

Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd 749 L. G. Cook and M. D. Crisp

Table 2 Maximum likelihood estimates of ancestral areas following re-analysis of data from Winkworth et al. (2002a) using a Markov continuous time model implemented in discrete

Hebe Ranunculus Myosotis

Rate low Rate high Rate low Rate high Rate low Rate high

Rate bias Basal Other Basal Other Basal Other Basal Other Basal Other Basal Other towards NZ node nodes node nodes node nodes node nodes node nodes node nodes

5 to 1 50 71–79 50 83 100 98–100 50 83–87 100 100 66 83 2 to 1 51 65–76 50 67 100 99–100 54 66–83 100 100 80 71 1 to 1 65 81–92 50 50 100 100 66 51–83 100 100 89 70 1 to 2 65 69–78 50 33 100 100 59 34–67 100 100 88 44 1to5 51 25–31 50 17 100 100 51 17–33 100 100 85 17

Values are preference for New Zealand (> 50) or Australia (< 50), expressed as a percentage. Preference for Australia is shown in bold. common than in the other direction, reconstructions for all This result seems paradoxical compared with the parsimony three of these taxa favour Australian ancestors for the New results and the expectation of the NAA scenario. The following Zealand clades. Notably, the change for hebes occurred with a is how we interpret what the Markovian model is doing. When directional bias of only 2 : 1. the instantaneous transition rates (dispersal rates) are high, multiple changes occur along each branch, and the estimates of ancestral states tend to reflect directly the relative rates, Likelihood irrespective of the terminal states. This can be clearly seen in multistate was used to model multiple areas in addition to the high-rates results for ‘other nodes’ in the rate-smoothed Australia and New Zealand, such as New Guinea and South hebe data (Table 2). When the dispersal rates are low, it is America. Although all three multistate data sets failed to unlikely that there will be change along any branch, and the meet the optimal ratio of terminal taxa to parameters (Pagel, reconstructions tend to reflect the predominant state among 2003), the results from multistate analysis did not differ terminal taxa. In all three data sets, New Zealand is by far the qualitatively from those of the discrete analysis, and only the most common terminal taxon and, in all three, this area is latter are presented here (Table 2). The terminal taxa from strongly favoured in reconstructions when rates are low areas other than Australia and New Zealand were pruned from (Table 2). A NAA scenario requires both (1) Australia to be the data sets used in discrete. For the hebe data set, we ancestral at the nodes along the backbone of the phylogeny, compared results using uncorrected branch lengths with those and (2) the terminal states to be predominantly New Zealand. using local rate smoothing (NPRS) as implemented in TreeEdit This implies lack of change (stasis) in states along the v1.0a10 (Rambaut & Charleston, 2002, http://evolve.zoo. backbone of the tree, with all the change (dispersals) occurring ox.ac.uk/software.html?id¼treeedit). The results were not in the terminal branches (with a strong bias towards New qualitatively different and only those from the rate-smoothed Zealand). The Markovian ML model appears to be unable to tree are reported here. model this differential pattern of evolution over the tree, and The Markovian ML model implemented in discrete was so cannot reconstruct a NAA scenario. This finding is similar unable to reconstruct a NAA scenario from any of the three to that shown by Oakley & Cunningham (2000) for data sets. Allowing the model to estimate the rates of dispersal continuous, rather than discrete, directionally biased traits. from the trees gave the same result for all three data sets: (1) a Of the three taxa discussed above, two species (Chionohebe higher rate in the direction of Australia and (2) a strong densifolia and Myosotis australis) allow the possibility of using preference for New Zealand as the ancestral area at most population genetics and phylogeography to attempt to internal nodes (not shown). Constraining ranges of rates for determine the direction of dispersal because they are both asymmetric and symmetric dispersal (Table 2) also gave a apparently shared between Australia and New Zealand. strong preference for New Zealand as the ancestral area at most Similarly, Hebe elliptica is shared between NZ and the Falkland internal nodes in most cases. The only constraints that led to Islands and could be used to assess direction of dispersal reconstructions preferring Australia as the ancestral area were between these two localities. the combination of high dispersal rates with a strong bias towards Australia (shown in bold in Table 2). This is the CONCLUSIONS opposite of the NAA scenario, which predicts a dispersal rate bias towards New Zealand, from an ancestral area in Australia. This paper demonstrates that there are plausible alternative However, when dispersal rates were constrained with a bias biogeographical reconstructions for taxon–area phylogenies towards New Zealand, NZ was always reconstructed as that are currently not being considered. This is mainly because ancestral, in all three data sets (Table 2). tree topology is given primacy over biogeographical processes.

750 Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd Asymmetrical dispersal and biogeography

Additionally, inferences about non-terminal speciation events Croizat, L., Nelson, G. & Rosen, D.E. (1974) Centers of origin (the cause of internal nodes in taxon phylogenies) that are and related concepts. Systematic Zoology, 23, 265–287. implicit in many reconstructions (e.g. sympatric speciation) Cunningham, C.W. (1999) Some limitations of ancestral are ignored. Consideration of all factors (phylogeny, process, character-state reconstruction when testing evolutionary speciation and fossils) is required to enhance our interpret- hypotheses. Systematic Biology, 48, 665–674. ation of biogeography. In particular, the effects of possibly Cunningham, C.W., Omland, K.E. & Oakley, T.H. (1998) asymmetrical dispersal should be considered. Reconstructing ancestral character states: a critical reappraisal. Trends in Ecology and Evolution, 13, 361–366. Dean, M.D. & Ballard, J.W.O. (2004) Linking phylogenetics ACKNOWLEDGEMENTS with population genetics to reconstruct the geographic We thank T.G. Barraclough, P.S. Cranston, P.J. Gullan, A.O. origin of a species. Molecular Phylogenetics and Evolution, 32, Mooers, R.D.M. Page, V. Savolainen, the Coopers and 998–1009. Cladistics discussion group (Canberra) and two anonymous Donoghue, M.J. & Moore, B.R. (2003) Toward an integrative reviewers for critical comments on drafts. historical biogeography. Integrative and Comparative Biol- ogy, 43, 261–270. Ebach, M.C. & Humphries, C.J. (2002) Cladistic biogeography REFERENCES and the art of discovery. Journal of Biogeography, 29, 427– Andersson, L. & Chase, M.W. (2001) Phylogeny and classiﬁ- 444. cation of Marantaceae. Botanical Journal of the Linnean Enghoff, H. (1995) Historical biogeography of the Holarctic: Society, 135, 275–287. area relationships, ancestral areas, and dispersal of non- Barraclough, T.G. & Nee, S. (2001) Phylogenetics and specia- marine animals. Cladistics, 11, 223–263. tion. Trends in Ecology and Evolution, 16, 391–399. Farris, J.S. (1983) The logical basis of phylogenetic analysis. Belshaw, R. & Quicke, D.L.J. (2002) Robustness of ancestral Advances in cladistics, vol. 2 (ed. by N.I. Platnick and V.A. state estimates: evolution of life history strategy in ichneu- Funk), pp. 7–36. Columbia University Press, New York. monoid parasitoids. Systematic Biology, 51, 450–477. Felsenstein, J. (2004) Inferring phylogenies. Sinauer Associates, Bowler, J.M. (1982) Aridity in the late Tertiary and Quaternary Sunderland. of Australia. Evolution of the ﬂora and fauna of arid Australia Fjeldsa, J. (1994) Geographical patterns for relict and young (ed. by W.R. Barker and P.J.M. Greenslade), pp. 35–45. species of birds in Africa and South America and implica- Peacock Publications, Glen Osmond. tions for conservation priorities. Biodiversity and Conserva- Bremer, K. (1992) Ancestral areas: a cladistic reinterpretation of tion, 3, 207–226. the center of origin concept. Systematic Biology, 41, 436–445. Gaines, S.D., Gaylord, B. & Largier, J.L. (2003) Avoiding Bush, M.B. & Whittaker, R.J. (1991) Krakatau: colonization current oversights in marine reserve design. Ecological patterns and hierarchies. Journal of Biogeography, 18, 341– Applications, 13, S32–S46. 356. Givnish, T.J. & Renner, S.S. (2004) Tropical intercontinental Calsbeek, R. & Smith, T.B. (2003) Ocean currents mediate disjunctions: Gondwana breakup, immigration from the evolution in island lizards. Nature, 426, 552–555. boreotropics, and transoceanic dispersal. International Campbell, G.S., Blackwell, P.G. & Woodward, F.I. (2002) Can Journal of Plant Sciences, 165, S1–S6. landscape-scale characteristics be used to predict plant Greathead, D.J. (1990) Crawler behaviour and dispersal. invasions along rivers? Journal of Biogeography, 29, 535–543. Armored scale insects: their biology, natural enemies and Colls, K. & Whitaker, R. (2001) The Australian weather book, control (ed. by D. Rosen), pp. 305–308. Elsevier Science, 2nd edn. Reed New Holland, Sydney. Amsterdam. Cook, L.G. & Gullan, P.J. (2004) The gall-inducing habit has Haddrath, O. & Baker, A.J. (2001) Complete mitochondrial evolved multiple times among the eriococcid scale insects DNA genome sequences of extinct birds: ratite phylogenetics (Sternorrhyncha: Coccoidea: Eriococcidae). Biological Jour- and the vicariance biogeography hypothesis. Proceedings of nal of the Linnean Society, 83, 441–452. the Royal Society of London B, Biological Sciences, 268, 939– Cooper, A., Lalueza-Fox, C., Anderson, S., Rambaut, A., 945. Austin, J. & Ward, R. (2001) Complete mitochondrial Hill, R.S. (1994) The history of selected Australian taxa. genome sequences of two extinct moas clarify ratite evolu- History of the Australian vegetation: Cretaceous to recent tion. Nature, 409, 704–707. (ed. by R.S. Hill), pp. 390–419. Cambridge University Press, Craw, R. (1982) Phylogenetics, areas, geology and the bioge- Cambridge. ography of Croizat: a radical view. Systematic Zoology, 31, Hill, R.S. (2001) Biogeography, evolution and palaeoecology of 304–316. Nothofagus (Nothofagaceae): the contribution of the fossil Craw, R. (1989) Continuing the synthesis between panbioge- record. Australian Journal of Botany, 49, 321–332. ography, phylogenetic systematics and geology as illustrated Howarth, D.G., Gustafsson, M.H.G., Baum, D.A. & Motley, by empirical studies on the biogeography of New Zealand T.J. (2003) Phylogenetics of the genus Scaevola (Good- and the Chatham Islands. Systematic Zoology, 37, 291–310. eniaceae): implication for dispersal patterns across the

Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd 751 L. G. Cook and M. D. Crisp

Pacific Basin and colonization of the Hawaiian Islands. Linder, H.P. & Crisp, M.D. (1995) Nothofagus and Pacific American Journal of Botany, 90, 915–923. biogeography. Cladistics, 11, 5–32. Huelsenbeck, J.P. & Bollback, J.P. (2001) Empirical and hier- Lockhart, P.J., McLenachan, P.A., Harell, D., Glenny, D., archical Bayesian estimation of ancestral states. Systematic Huson, D. & Jensen, U. (2001) Phylogeny, radiation, and Biology, 50, 351–366. transoceanic dispersal of New Zealand alpine buttercups: Huelsenbeck, J.P., Larget, B., Miller, R.E. & Ronquist, F. (2002) molecular evidence under split decomposition. Annals of the Potential applications and pitfalls of Bayesian inference of Missouri Botanical Garden, 88, 458–477. phylogeny. Systematic Biology, 51, 673–688. Lutzoni, F., Pagel, M. & Reeb, V. (2001) Major fungal lineages Huelsenbeck, J.P., Nielsen, R. & Bollback, J.P. (2003) Sto- are derived from lichen symbiotic ancestors. Nature, 411, chastic mapping of morphological characters. Systematic 937–940. Biology, 52, 131–158. Macphail, M.K. (1997) The New Zealand flora: entirely long- Humphries, C.J. (2001) Vicariance biogeography. Encyclopedia distance dispersal? Comment. Journal of Biogeography, 24, of biodiversity, Vol. 5 (ed. by S. Levin), pp. 767–779. Aca- 113–117. demic Press, San Diego. Maddison, W.P. & Maddison, D.R. (1992) MacClade: analysis Hunn, C.A. & Upchurch, P. (2001) The importance of time/ of phylogeny and character evolution, ver. 3.0. Sinauer space in diagnosing the causality of phylogenetic events: Associates, Sunderland. towards a ‘chronobiogeographical’ paradigm? Systematic Manos, P.S. & Stanford, A.M. (2001) The historical bio- Biology, 50, 391–407. geography of Fagaceae: tracking the Tertiary history of Jordan, G.J. (2001) An investigation of long-distance dispersal temperate and subtropical forests of the Northern based on species native to both Tasmania and New Zealand. Hemisphere. International Journal of Plant Sciences, 162, Australian Journal of Botany, 49, 333–340. S77–S93. Kroenke, L.W. (1996) Plate tectonic development of the Mast, A.R. & Givnish, T.J. (2002) Historical biogeography and western and southwestern Pacific: Mesozoic to the present. the origin of stomatal distributions in Banksia and Dryandra The origin and evolution of Pacific island biotas, New Guinea (Proteaceae) based on their cpDNA phylogeny. American to eastern Polynesia: patterns and processes (ed. by A. Keast Journal of Botany, 89, 1311–1323. and S.E. Miller), pp. 19–34. SPB Academic Publishing, McCarthy, D. (2003) The trans-Pacific zipper effect: disjunct Amsterdam. sister taxa and matching geological outlines that link the Lee, D.E., Lee, W.G. & Mortimer, N. (2001) Where and why Pacific margins. Journal of Biogeography, 30, 1545–1561. have all the flowers gone? Depletion and turnover in the McDowall, R.M. (2002) Accumulating evidence for a dispersal New Zealand Cenozoic angiosperm flora in relation to biogeography of southern cool temperate freshwater fishes. palaeogeography and climate. Australian Journal of Botany, Journal of Biogeography, 29, 207–219. 49, 341–356. McDowall, R.M. (2004) What biogeography is: a place for Les, D.H., Crawford, D.J., Kimball, R.T., Moody, M.L. & process. Journal of Biogeography, 31, 345–351. Landolt, E. (2003) Biogeography of discontinuously dis- McGlone, M.S., Duncan, R.P. & Heenan, P.B. (2001) tributed hydrophytes: a molecular appraisal of inter- Endemism, species selection and the origin and distribu- continental disjunctions. International Journal of Plant tion of the vascular plant flora of New Zealand. Journal of Sciences, 164, 917–932. Biogeography, 28, 199–216. Levin, S.A., Muller-Landau, H.C., Nathan, R. & Chave, J. Mooers, A.O. & Schluter, D. (1999) Reconstructing ancestor (2003) The ecology and evolution of seed dispersal: a the- states with maximum likelihood: support for one- and two- oretical perspective. Annual Review of Ecology Evolution and rate models. Systematic Biology, 48, 623–633. Systematics, 34, 575–604. Mummenhoff, K., Linder, P., Friesen, N., Bowman, J.L., Lee, Levine, J.M. (2003) A patch modeling approach to the com- J.Y. & Franzke, A. (2004) Molecular evidence for munity-level consequences of directional dispersal. Ecology, bicontinental hybridogenous genomic constitution in 84, 1215–1224. Lepidium sensu stricto (Brassicaceae) species from Aus- Leys, R., Cooper, S.J.B. & Schwarz, M.P. (2000) Molecular phy- tralia and New Zealand. American Journal of Botany, 91, logeny of the large carpenter bees, genus Xylocopa (Hyme- 254–261. noptera: Apidae), based on mitochondrial DNA sequences. Munõz, J., Felicisimo, A.M., Cabezas, F., Burgaz, A.R. & Molecular Phylogenetics and Evolution, 17, 407–418. Martinez, I. (2004) Wind as a long-distance dispersal Lieberman, B.S. (2002) Phylogenetic analysis of some basal vehicle in the Southern Hemisphere. Science, 304, 1144– early Cambrian trilobites, the biogeographic origins of the 1147. Eutrilobita, and the timing of the Cambrian radiation. Nelson, G. & Ladiges, P.Y. (1996) Paralogy in cladistic bio- Journal of Paleontology, 76, 692–708. geography and analysis of paralogy-free subtrees. American Linder, H.P. (1995) Setting conservation priorities: the Museum Novitiates, 3167, 1–58. importance of endemism and phylogeny in the southern Nelson, G. & Ladiges, P.Y. (2001) Gondwana, vicariance bio- African orchid genus Herschelia. Conservation Biology, 9, geography and the New York School revisited. Australian 585–595. Journal of Botany, 49, 389–409.

752 Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd Asymmetrical dispersal and biogeography

Nelson, G. & Platnick, N.I. (1981) Systematics and biogeogra- Rambaut, A. & Charleston, M.A. (2002) TreeEdit: phylogenetic phy: cladistics and vicariance. Columbia University Press, tree editor, ver. 1.0 alpha 10. Department of Zoology, Uni- New York. versity of Oxford, Oxford. Nepokroeff, M., Sytsma, K.J., Wagner, W.L. & Zimmer, E.A. Ranker, T.A., Floyd, S.K. & Trapp, P.G. (1994) Multiple (2003) Reconstructing ancestral patterns of colonization and colonizations of Asplenium adiantum-nigrum onto the dispersal in the Hawaiian understorey tree genus Psychotria Hawaiian archipelago. Evolution, 48, 1364–1370. (Rubiaceae): a comparison of parsimony and likelihood Raven, P.H. (1973) Evolution of subalpine and alpine plant approaches. Systematic Biology, 52, 820–838. groups in New Zealand. New Zealand Journal of Botany, 11, Nielsen, R. (2002) Mapping mutations on phylogenies. Sys- 177–200. tematic Biology, 51, 729–739. Ree, R.H. & Donoghue, M.J. (1998) Step matrices and the Oakley, T.H. & Cunningham, C.W. (2000) Independent con- interpretation of homoplasy. Systematic Biology, 47, 582–588. trasts succeed where ancestor reconstruction fails in a Ree, R.H. & Donoghue, M.J. (1999) Inferring rates of change known bacteriophage phylogeny. Evolution, 54, 397–405. in flower symmetry in Asterid angiosperms. Systematic Oakley, T.H. & Cunningham, C.W. (2002) Molecular phylo- Biology, 48, 633–641. genetic evidence for the independent evolutionary origin of Renner, S.S., Foreman, D. & Murray, D. (2000) Timing an arthropod compound eye. Proceedings of the National transantarctic disjunctions in the Atherospermataceae Academy of Sciences of the United States of America, 99, (Laurales): evidence from coding and noncoding chloroplast 1426–1430. sequences. Systematic Biology, 49, 579–591. Omland, K.E. (1997) Examining two standard assumptions of Renner, S.S., Clausing, G. & Meyer, K. (2001) Historical ancestral reconstructions: repeated loss of dichromatism in biogeography of Melastomataceae: the roles of Tertiary dabbling ducks (Anatini). Evolution, 51, 1636–1646. migration and long-distance dispersal. American Journal of Omland, K.E. (1999) The assumptions and challenges of Botany, 88, 1290–1300. ancestral state reconstructions. Systematic Biology, 48, 604– Ronquist, F. (1994) Ancestral areas and parsimony. Systematic 611. Biology, 43, 267–274. Page, R.D.M. (1994) Maps between trees and cladistic analysis Ronquist, F. (1997) Dispersal-vicariance analysis: a new of historical associations among genes, organisms, and areas. approach to the quantification of historical biogeography. Systematic Biology, 43, 58–77. Systematic Biology, 46, 195–203. Page, R.D.M. (2003) Introduction. Tangled trees: phylogeny, Ronquist, F. (1998) Phylogenetic approaches in coevolution cospeciation, and coevolution (ed. by R.D.M. Page), pp. 1–21. and biogeography. Zoologica Scripta, 26, 313–322. University of Chicago Press, Chicago and London. Ronquist, F. (2003) Parsimony analysis of coevolving species Pagel, M. (1994) Detecting correlated evolution on phyloge- associations. Tangled trees: phylogeny, cospeciation and nies: a general method for the comparative analysis of dis- coevolution (ed. by R.D.M. Page), pp. 22–64. University of crete characters. Proceedings of the Royal Society of London B, Chicago Press, Chicago. Biological Sciences, 255, 37–45. Ronquist, F (2004) Bayesian inference of character evolution. Pagel, M. (1999a) The maximum likelihood approach to Trends in Ecology and Evolution, 19, 475–481. reconstructing ancestral character states of discrete charac- Ronquist, F. & Nylin, S. (1990) Process and pattern in the ters on phylogenies. Systematic Biology, 48, 612–622. evolution of species associations. Systematic Zoology, 39, Pagel, M. (1999b) Inferring the historical patterns of biological 323–344. evolution. Nature, 401, 877–884. Sanmartıń, I. & Ronquist, F. (2004) Southern Hemisphere Pagel, M. (2000) DISCRETE 4.0: correlated evolution, ancestral biogeography inferred by event-based models: plant versus states and trait evolution. University of Reading, Reading. animal patterns. Systematic Biology, 53, 216–243. Pagel, M. (2003) MULTISTATE 0.8. University of Reading, Schluter, D., Price, T., Mooers, A.O. & Ludwig, D. (1997) Reading. Likelihood of ancestor states in adaptive radiation. Evolu- Pagel, M., Meade, A. & Barker, D. (2004) Bayesian estimation tion, 51, 1699–1711. of ancestral character states on phylogenies. Systematic Schluter, D., Boughman, J.W. & Rundle, H.D. (2001) Parallel Biology, 53, 758–766. speciation with allopatry. Trends in Ecology and Evolution, Pedgley, D.E. (1982) Windborne pests and diseases: meteorology 16, 283–284. of airborne organisms. Ellis Horwood Ltd, New York. Schultz, T.R. & Churchill, G.A. (1999) The role of subjectivity Pole, M.S. (1994) The New Zealand flora: entirely long-dis- in reconstructing ancestral character states: a Bayesian tance dispersal? Journal of Biogeography, 21, 625–635. approach to unknown rates, states, and transformation Pole, M.S. (2001) Can long-distance dispersal be inferred from asymmetries. Systematic Biology, 48, 651–664. the New Zealand plant fossil record? Australian Journal of Slikas, B., Olson, S.L. & Fleischer, R.C. (2002) Rapid, Botany, 49, 357–366. independent evolution of flightlessness in four species of Rambaut, A. (1999) TreeThief: a tool for manual phylogenetic Pacific Island rails (Rallidae): an analysis based on tree entry from scanned images. Department of Zoology, mitochondrial sequence data. Journal of Avian Biology, 33, University of Oxford, Oxford. 5–14.

Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd 753 L. G. Cook and M. D. Crisp

Soltis, P.S. & Gitzendanner, M.A. (1999) Molecular systematics widespread African-Malagasy frog lineage. Journal of Bio- and the conservation of rare species. Conservation Biology, geography, 31, 593–601. 13, 471–483. Wagner, W.L. & Funk, V.A. (Eds) (1995) Hawaiian biogeo- Swenson, U. & Bremer, K. (1997) Pacific biogeography of the graphy: evolution on a hot spot archipelago. Smithsonian Asteraceae genus Abrotanella (Senecioneae, Blennosperma- Institution Press, Washington. tinae). Systematic Botany, 22, 493–508. Wagner, D.L. & Liebherr, J.K. (1992) Flightlessness in insects. Swenson, U., Backlund, A., McLoughlin, S. & Hill, R.S. (2001) Trends in Ecology and Evolution, 7, 216–220. Nothofagus biogeography revisited with special emphasis on Wagstaff, S.J., Bayly, M.J., Garnock-Jones, P.J. & Albach, D.C. the enigmatic distribution of subgenus Brassospora in New (2002) Classification, origin, and diversification of the New Caledonia. Cladistics, 17, 28–47. Zealand hebes (Scrophulariaceae). Annals of the Missouri Swofford, D.L. (2002) PAUP*: phylogenetic analysis using par- Botanical Garden, 89, 38–63. simony (*and other methods), ver. 4b10. Sinauer Associates, Wardle, P. (1978) Origin of the New Zealand mountain flora, Sunderland. with special reference to trans-Tasman relationships. New Taylor, E.B. & McPhail, J.D. (2000) Historical contingency and Zealand Journal of Botany, 16, 535–550. ecological determinism interact to prime speciation in Wares, J.P., Gaines, S.D. & Cunningham, C.W. (2001) A sticklebacks, Gasterosteus. Proceedings of the Royal Society of comparative study of asymmetric migration events across a London B, Biological Sciences, 267, 2375–2384. marine biogeographic boundary. Evolution, 55, 295–306. Taylor, E.B., McPhail, J.D. & Schluter, D. (1997) History of Waters, J.M. & Roy, M.S. (2004) Out of Africa: the slow train ecological selection in sticklebacks: uniting experimental and to Australasia. Systematic Biology, 53, 18–24. phylogenetic approaches. Molecular evolution and adaptive Waters, J.M., Lopez, J.A. & Wallis, G.P. (2000a) Molecular radiation (ed. by T.J. Givnish and K.J. Sytsma), pp. 511–534. phylogenetics and biogeography of galaxiid fishes Cambridge University Press, Cambridge. (Osteichthyes: Galaxiidae): dispersal, vicariance, and the Templeton, A.R. (1998) Nested clade analyses of phylogeo- position of Lepidogalaxias salamandroides. Systematic Biol- graphic data: testing hypotheses about gene flow and pop- ogy, 49, 777–795. ulation history. Molecular Ecology, 7, 381–397. Waters, J.M., Dijkstra, L.H. & Wallis, G.P. (2000b) Thorne, R. (2004) Tropical plant disjunctions: a personal Biogeography of a southern hemisphere freshwater fish: how reflection. International Journal of Plant Science, 165(4 important is marine dispersal? Molecular Ecology, 9, Suppl.), S137–S138. 1815–1821. Trewick, S.A. (1997) Flightlessness and phylogeny amongst Wenny, D.G. & Levey, D.J. (1998) Directed seed dispersal by endemic rails (Aves, Rallidae) of the New Zealand region. bellbirds in a tropical cloud forest. Proceedings of the Philosophical Transactions of the Royal Society London B, National Academy of Sciences of the United States of America, Biological Sciences, 352, 429–446. 95, 6204–6207. Trewick, S.A. (2000) Molecular evidence for dispersal rather Whiting, M.F., Bradler, S. & Maxwell, T. (2003) Loss and than vicariance as the origin of flightless insect species on recovery of wings in stick insects. Nature, 421, 264–267. the Chatham Islands, New Zealand. Journal of Biogeography, Winkworth, R.C., Wagstaff, S.J., Glenny, D. & Lockhart, P.J. 27, 1189–1200. (2002a) Plant dispersal N.E.W.S. from New Zealand. Trends Turelli, M., Barton, N.H. & Coyne, J.A. (2001) Theory and in Ecology and Evolution, 17, 514–520. speciation. Trends in Ecology and Evolution, 16, 330–343. Winkworth, R.C., Grau, J., Robertson, A.W. & Lockhart, P.J. Veevers, J.J., Powell, C.M. & Roots, S.R. (1991) Review of sea- (2002b) The origins and evolution of the genus Myosotis L. floor spreading around Australia. 1. Synthesis of the patterns (Boraginaceae). Molecular Phylogenetics and Evolution, 24, of spreading. Australian Journal of Earth Sciences, 38, 373– 180–193. 389. Wolf, P.G., Schneider, H. & Ranker, T.A. (2001) Geographic Vences, M., Kosuch, J., Rodel, M.O., Lotters, S., Channing, A., distributions of homosporous ferns: does dispersal obscure Glaw, F. & Bohme, W. (2004) Phylogeography of Ptycha- evidence of vicariance? Journal of Biogeography, 28, 263– dena mascareniensis suggests transoceanic dispersal in a 270.

BIOSKETCHES

Lyn Cook has interests in speciation, molecular phylogenetics, biogeography, karyology and scale insect–plant interactions.

Mike Crisp has an interest in phylogenies and their application to questions in macro-evolution, biogeography, ecology, speciation and the evolution of adaptive traits.

Editor: David Bellwood

754 Journal of Biogeography 32, 741–754, ª 2005 Blackwell Publishing Ltd